Googleโs CEO says AI is jagged, not general.
Hereโs what that means:
โข ๐ช๐ต๐ฒ๐๐ต๐ฒ๐ฟ ๐๐๐ ๐ฎ๐ฟ๐ฟ๐ถ๐๐ฒ๐ ๐ผ๐ฟ ๐ป๐ผ๐, ๐๐ฒโ๐น๐น ๐๐๐ถ๐น๐น ๐ณ๐ฎ๐ฐ๐ฒ ๐บ๐ฎ๐๐๐ถ๐๐ฒ ๐ฑ๐ถ๐๐ฟ๐๐ฝ๐๐ถ๐ผ๐ป: Token use is exploding. Real use cases are forming. Societyโs shockwaves are already here.
โข ๐๐ ๐ฎ๐ฑ๐๐ฎ๐ป๐ฐ๐ฒ๐ ๐ฎ๐ฟ๐ฒ ๐ฝ๐ฎ๐๐ฐ๐ต๐โ๐ฏ๐ฟ๐ถ๐น๐น๐ถ๐ฎ๐ป๐ฐ๐ฒ ๐บ๐ถ๐
๐ฒ๐ฑ ๐๐ถ๐๐ต ๐ฏ๐น๐๐ป๐ฑ๐ฒ๐ฟ๐: Sundar Pichai calls this AJIโArtificial Jagged Intelligence. Weโll see stunning breakthroughs alongside mistakes. That unpredictability is now the default.
โข ๐๐ ๐ถ๐ ๐บ๐ผ๐ฟ๐ฒ ๐ฝ๐ฟ๐ผ๐ณ๐ผ๐๐ป๐ฑ ๐๐ต๐ฎ๐ป ๐ณ๐ถ๐ฟ๐ฒ ๐ผ๐ฟ ๐ฒ๐น๐ฒ๐ฐ๐๐ฟ๐ถ๐ฐ๐ถ๐๐: The ripple effects will reshape education, jobs, politics, and cultureโoften in ways we canโt model. Think second- and third-order impacts, not just new apps.
โข ๐๐ฎ๐๐ฒ๐ป๐ฐ๐ ๐ฏ๐ฒ๐ฎ๐๐ ๐ถ๐ป๐๐ฒ๐น๐น๐ถ๐ด๐ฒ๐ป๐ฐ๐ฒ: In real-world usage, a dumber but faster model often wins. If it responds instantly, it feels smarter. Founders should optimize for speed.
โข ๐๐ฒ๐ป๐ฐ๐ต๐บ๐ฎ๐ฟ๐ธ๐ ๐ฎ๐ฟ๐ฒ ๐ผ๐๐โ๐ผ๐๐๐ฐ๐ผ๐บ๐ฒ๐ ๐ฎ๐ฟ๐ฒ ๐ถ๐ป: Users donโt care if your model wins on LMSYS or MMLU. They care if it helps them write, think, decide, or create better. Founders should focus on usefulness, not leaderboard rankings.
โข ๐๐ ๐ถ๐ ๐ฏ๐ฒ๐ฐ๐ผ๐บ๐ถ๐ป๐ด ๐ฎ๐บ๐ฏ๐ถ๐ฒ๐ป๐ ๐ถ๐ป๐ณ๐ฟ๐ฎ: Like Google Docs, future apps wonโt announce their intelligence. Theyโll just work. Invisible, fast, helpful. AI melts into the interface.
๐ Source: clip from Vitrupo on X, original source Sundar Pichai Lex Fridman Podcastย #471 on Youtube.